Btw, I don't think there is any undocumented register, or even bit, in GTIA or in any of the chipset at all. What you have is a couple of things not 100% fully documented, bugs like the one in Antic, and some tricks like Bryan's method for shifting GTIA modes.
I would expect there to be lots of undocumented registers, but for them not to be memory mapped; most of them will typically work together to mimic the behavior of documented "registers" that don't actually exist in the documented form.
I'm not familiar with the CTIA/GTIA, but to borrow some examples from the original TIA...
-1- The documentation shows that each sprite has a divide-by-160 counter. The documentation does not show that each sprite actually has a cascaded divide-by-four and a divide-by-forty. Usually, the cascaded counters behave like a divide-by-160, but the synchronization logic gets tripped up if a sprite is reset within a small window of its normal trigger point.
-2- The documentation states that there is a pulser circuit which adds and subtracts pulses. It does not mention that the subtraction of pulses is achieved by delaying the end of hblank, nor does it mention that there is a four-bit counter that counts once every four pixel clocks except that if it's sitting at zero it will remain there until the next HMOVE. The documentation also does not mention that each sprite has a latch which turns its pulser on and off, and that this latch may be 'jammed'.
-3- The documentation does not describe the particular behavior of the shift registers used in the audio generation circuitry. Though it accurately describes what any particular AUDCx mode will do in isolation, the description is not sufficient to know what will happen if AUDCx is changed while a sound is playing.
All of those behaviors are the result of incompletely-documented registers. I'm sure many more examples could be found for the TIA.